The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of a web server application that processes untrusted input from external users via HTTP request handlers.
Ingestion points: Data enters the application context through UsersController and CatsController using decorators like @Body(), @Param(), and @Query() in SKILL.md and reference.md.
Boundary markers: The provided code does not demonstrate the use of specific boundary delimiters or instructions to prevent the model from obeying data-embedded commands, which is typical for general application templates.
Capability inventory: The skill provides database interaction capabilities (select, insert, update, delete) through Drizzle ORM and TypeORM, and file/configuration access through dotenv.
Sanitization: The skill explicitly includes a ValidationPipe implementation using class-validator in SKILL.md to ensure external input adheres to defined schemas before processing.
[CREDENTIALS_UNSAFE]: Configuration examples in the skill include hardcoded fallback strings for sensitive security tokens.
Evidence: SKILL.md uses secret: process.env.JWT_SECRET || 'default-secret'. Providing a static fallback for a JWT secret can lead to insecure deployments if a user fails to configure the corresponding environment variable.
[EXTERNAL_DOWNLOADS]: The drizzle-reference.md file contains numerous commands to download and install packages from public registries.
Evidence: The file instructs users to install drizzle-orm, @planetscale/database, @neondatabase/serverless, and other drivers. These are standard packages from well-known technology providers for the described use case.
[COMMAND_EXECUTION]: The skill provides shell commands for environment configuration and local testing.
Evidence: drizzle-reference.md includes docker run commands for starting database instances and turso auth commands for CLI-based service authentication.

nestjs